Unitxt 1.5.0
What's Changed
- Add Perplexity by @assaftibm in #442
- Balance demos in fair_tos datasets by @yoavkatz in #473
- postprocessors become extensions of FieldOperator by @dafnapension in #468
- Initial UI by @Roni-Friedman in #475
- introducing text at the top of loaders.py by @dafnapension in #474
- Improve Unitxt UI and add support for multiple catalogs by @Roni-Friedman in #476
- Improve ExtractMostCommonFieldValues performance by avoiding the creation of the stream-length long list of values instead building values counter on the fly by @dafnapension in #471
- Add a new Perturbate operator that is handy for faking prediction as a perturbated version of target or for any other perturbation use cases by @dafnapension in #456
- add a card for HF xsum, a summary dataset by @dafnapension in #479
- Add QA, NER, Targeted sentiment and Generation tasks, Llama and Alpaca formats and instructions and whitespace augmentor by @matanor in #483
- Cfpb product by @ilyashnil in #485
- Improve catalog UI on documentation website by separating catalog objects to files and adding information per catalog item by @matanor in #461
- Update QA templates by @matanor in #486
- Length balancer docstring by @matanor in #487
- Add process_instance function to every mutli stream operator for easier testing and usage at instance level by @elronbandel in #488
- Add LRU caching for catalog artifact loading to minimize IO overhead and enhance performance by @elronbandel in #489
- Make postprocessors a general operator that operates on 'prediction' and on 'references' which enables the use of every possible operator as postprocessor by @dafnapension in #484
- Improve UI code presentation and organization by @Roni-Friedman in #491
- Simplify ui launching with the console command unitxt-explore by @elronbandel in #492
- Xmmlu template multilingual by @gitMichal in #493
New Contributors
- @Roni-Friedman made their first contribution in #475
Full Changelog: 1.4.6...1.5.0